This proposal suggests a model and a series of experiments to explain the polarization of attitudes following group discussion in terms of the production and dissemination of cogent information by individual members. The model assumes that the person who is about to indicate a preference generates a sample of persuasive arguments (ideas, thoughts, images describing positive and negative features of each alternative) from a larger culturally-given pool of arguments. The attitude expressed will depend on the balance of arguments in the sample. Changes in attitude occur when new arguments, ones which are available but not readily generated by the person, enter the sample. The model describes several ways in which this can happen. For instance, polarization toward alternative J may result when either (1) some members are induced to generate additional pro-J arguments which are persuasive but unrepresentative of the larger pool and thus unlikely to come to mind ordinarily, or (2) the group directly provides such pro-J arguments to these members. Thus, when the distribution of arguments in the larger pool and in each member's sample are known, the magnitude and direction of polarization can be predicted with uncommon precision. Finally, it is suggested that the persuasive-arguments formulation may account for a wide range of social influence phenomena including several well known but puzzling effects in the classical conformity literature (e.g., a large unanimous majority has no more impact than a small one) and thus can have very general theoretical significance.